Scholarlink

<p>Rapid and accurate recognition of barcode information on express waybills is crucial for intelligent sorting in express delivery. Current algorithms for express waybill barcode detection face challenges, as high-precision models are too large to be deployed on all scanning devices, while lightweight models often lack detection accuracy. To address this, a novel YOLOv8-based network is proposed, balancing model size and accuracy. First, the Star Blocks from StarNet are integrated into the C2f module to form the C2f-Star module, reducing model parameters. Adaptive Fine-Grained Channel Attention (FCA) is then embedded into the C2f-Star module to create the FS-C2f module, achieving more efficient feature weighting. Next, a lightweight anti-aliasing Wavelet Pooling (WaveletPool) is introduced to replace downsampling, enhancing small-object detection accuracy. Lastly, a Lightweight Shared Convolution Head (LSHead) is proposed to further reduce parameters and improve detection precision. The improved model, named FWL-YOLO after its key enhancement modules, achieves MAP50 and MAP50:95 scores of 99.3% and 82.9% on the test set—representing improvements of 0.1% and 3.1% over YOLOv8s, respectively. Moreover, FWL-YOLO reduces parameter count and computational load by 39.5% and 37.1% respectively, with values of 6.74&#xa0;M and 18.0 GFLOPs. Thus, FWL-YOLO provides a promising lightweight solution for express waybill barcode detection, combining model compactness with enhanced detection accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FWL-YOLO: a lightweight model for barcode detection and recognition in express delivery waybills

  • 期刊论文

  • 发表日期：

    2025年1月28日

查看全文
  • Yuanhao Qu,
  • Fengshou Zhang

摘要

Rapid and accurate recognition of barcode information on express waybills is crucial for intelligent sorting in express delivery. Current algorithms for express waybill barcode detection face challenges, as high-precision models are too large to be deployed on all scanning devices, while lightweight models often lack detection accuracy. To address this, a novel YOLOv8-based network is proposed, balancing model size and accuracy. First, the Star Blocks from StarNet are integrated into the C2f module to form the C2f-Star module, reducing model parameters. Adaptive Fine-Grained Channel Attention (FCA) is then embedded into the C2f-Star module to create the FS-C2f module, achieving more efficient feature weighting. Next, a lightweight anti-aliasing Wavelet Pooling (WaveletPool) is introduced to replace downsampling, enhancing small-object detection accuracy. Lastly, a Lightweight Shared Convolution Head (LSHead) is proposed to further reduce parameters and improve detection precision. The improved model, named FWL-YOLO after its key enhancement modules, achieves MAP50 and MAP50:95 scores of 99.3% and 82.9% on the test set—representing improvements of 0.1% and 3.1% over YOLOv8s, respectively. Moreover, FWL-YOLO reduces parameter count and computational load by 39.5% and 37.1% respectively, with values of 6.74 M and 18.0 GFLOPs. Thus, FWL-YOLO provides a promising lightweight solution for express waybill barcode detection, combining model compactness with enhanced detection accuracy.